A Deep Learning Based Approach For Path Planning Of Mobile Charging In Wireless Rechargeable Sensor Network

With the continuous progress of microelectronics technology,wireless communication technology and embedded systems,wireless sensor networks(WSNs)technology has made great progress.WSNs are composed of many sensor nodes with sensing function,communication function and computing function.Sensor nodes need to consume power and energy to perform related tasks,such as sensing the surrounding environment and collecting environmental information.WSNs are widely used in traffic monitoring,environmental monitoring,smart home,medical treatment,military,and other fields.In traditional sensor networks,sensor nodes are deployed in a fixed location and supply power through battery equipment or through ambient energy,but such power supply mode is unstable and uncontrollable.In order to solve the energy supply problem of traditional wireless sensor networks,mobile charging devices are introduced to provide power to sensor nodes,which becomes an effective solution.Because there are many nodes in WSNs,their location,power status and demand for power supplement are different.Therefore,in order to enable the charging trolley to reasonably select the sensor nodes that need to be charged,balance the energy loss between nodes,perform the charging task effectively,and avoid the ”death” of nodes due to insufficient power in the network,an effective method is needed to carry out reasonable path planning for the mobile charging car.In the view of the above situation,this thesis proposes a method of deep reinforcement learning based on graph neural network to solve the problem of mobile charging device path planning.The main contributions of this thesis are summarized as follows:(1)Towards the mobile charging problem in wireless sensor networks,we present the problem definition of scene modeling,and determine the optimization goal to improve the rescue rate of the mobile charging device to the nodes and reduce the performance cost of the mobile charging device.(2)In order to solve the problem of charging path planning,this thesis proposes a deep learning network model algorithm which is combined with graph neural network,and uses the reinforcement learning actor critic method to train the model.This solves the problem of lack of optimal solution label in the data set.The graph and node information of the sensor network are processed,and the priority of node access is described by using the attention mechanism,so as to provide a reasonable charging strategy for the mobile charging device.(3)When the number of nodes in the sensor network increases,it is necessary to deploy multiple mobile charging device to the sensor network to ensure a high node survival rate.Under the scenario of multi mobile charging device,this thesis proposes a clustering algorithm,which divides the total charging task into several sub tasks,the number of which is equal to the number of devices,and then designs their own charging strategy for each mobile charging device through the network model proposed in(2).(4)Most existing algorithm models often take a long time to obtain the optimal solution when solving the traveling salesman problem(TSP).In this thesis,the generalization of the proposed network model is explored.The model is firstly trained on small-scale problems,such as TSP50,and then verified on large-scale problems.The experimental results show that the algorithm model proposed in this thesis can not be better than some traditional algorithms and standard solvers,but the solution speed is greatly improved,and a good suboptimal solution can be obtained in a reasonable solution time,which meets the needs of the actual scene,and verifies the generalization performance of the model.